Prognostics and health management method for aging systems

1. A prognostic and health management method for monitoring an aging system, including a component of said system, said method useful for determining whether or not detected anomalies are due to natural aging or other aging processes, said method based on a combination of a conventional data-driven method and a moving window data-driven method, including the steps of: detecting anomalies of said system, or component of said system based on monitored parameters of the system using both the conventional data-driven method and the moving window data-driven method; and identifying whether said detected anomalies arc due to natural aging or other aging processes or causes wherein; if both methods do not detect any anomalies, the system or component is healthy; if the conventional data-driven method detects anomalies but the moving window method does not, the detected anomalies are due to natural aging; if both methods detect anomalies, the detected anomalies are caused by other aging processes besides natural aging; wherein said detecting and identifying steps are performed with the aid of a computer.

2. The prognostic and health management method of claim 1 , wherein both the conventional and moving window data-driven method can be any data-driven method in which training data is used to train algorithms or to create a detection baseline or criteria, including the multivariate state estimation technique (MSET), neural network (NN), and Mahalanohis distance (MD) method.

3. The prognostic and health management method of claim 2 wherein the training data used in the training process is progressively updated by a moving window, and wherein a first training window is defined by a first data set collected over a first interval of time, the first data set representative of a healthy condition, a second, test window then defined by a test data set collected over a second interval of time, and the training data of the first window compared to the test data set of the second window to detect anomalies.

4. The prognostic and health management method of claim 3 wherein the moving window method employs either a single-sided moving window data-driven method, or a double-sided moving window data-driven method, size of the window or interval of time being fixed or flexible based on a particular application.

5. The prognostic and health management method of claim 4 employing a single-sided moving window wherein a first window is defined by a first data set collected over a first interval of time, the first data set representative of a healthy condition and used as training data, a second window then defined by a second data set collected over a second interval of time, the data set of the first window compared to the data set of the second window, whereby, if the data set of the second window is determined to be healthy, the second data set is added into the first window to expand the first data set, and the expanded data set used as new training data which is then compared with a test data set collected during a next interval of time to detect anomalies.

6. The prognostic and health management method of claim 4 , employing a double-side moving window wherein, a first training window is defined by a first data set collected over a first interval of time, the first data set representative of a healthy condition and used as training data, a second window then defined by a second data set collected over a second interval of time, the window of the first data set compared to the window of the second data set, whereby, if the data set of the second window is determined to be healthy, the data set of the second window is used as new training data for the next comparison, and the data set of the first window discarded, the new training data compared with a data set of a next interval of time to detect anomalies.

7. The prognostic and health management method of claim 3 , further including a simple threshold detection method to define anomalies, which method creates the criteria of a healthy system in terms of features of the training data and compares the features of the test data with the criteria.

8. The prognostic and health management method of claim 3 , further including a sequential probability ratio test or another decision-making method such as that can detect the anomalies based on features of collected data.

9. The prognostic and health management method of claim 3 wherein an anomaly is defined as at least five continuous test data readings that are outside of a defined healthy boundary.

10. The prognostic and health management method of claim 3 , further including a conventional multivariate state estimation technique (MSET) and a moving window-multivariate state estimation technique (MW-MSET) such that: if both methods do not detect anomalies, the system is deemed healthy; if the conventional MSET detects anomalies but the MW-MSET does not, then the detected anomalies are deemed to be due to natural aging; if both methods detect anomalies, the detected anomaly is anomalies are deemed to be caused by other aging processes besides natural aging.

11. The prognostic and health management method of claim 10 , wherein the performance of the conventional MSET for aging systems is improved by incorporating a moving window method in which training data is updated by an extending window, or a double-side moving window, for both of which size of the moving window or the amount of data used to update the training data is fixed or flexible.

12. The prognostic and health management method of claim 10 wherein either a simple threshold detection method, which creates the criteria of a healthy system in terms of calculated residuals from the training process and compares the calculated residuals of the test data with the criteria, or other decision-making method, such as a sequential probability ratio lest method, is used to detect anomalies based on the calculated residuals.